1. Field of the Invention
The invention is directed generally to compensating for undesirable characteristics in electronic signals, and, more particularly, to a method and device to compensate for baseline wander.
2. Related Art
In the past, baseline wander has not been a critical factor in the design of compensation systems. However, with the advancement of technology and more sensitive applications, baseline wander has become problematic. In particular, in many devices there is an idealized perpendicular channel response that has a strong DC component. In particular, there is a sharp notch at DC due to constraints imposed by real-world physics. Additionally, in such devices, for example, a readback signal may be coupled to a pre-amplifier or a channel through a high pass filter (HPF) to facilitate analog circuit design. However, this configuration leads to an undesirable long narrow tail in the pulse response. This tail takes substantially the form of an exponential decay. If the data pattern has significant build up of this DC component, a baseline shift can result. This baseline shift of the pulse response needs to be zero or near zero or errors in the system may be caused.
There have been three different approaches that have been taken for baseline wander compensation. These methods include: (1) error feedback loop (also known as DC-loop or adaptive loop); (2) linear equalization (also known as feed-forward equalization); and (3) decision feedback equalization (a form of the predictive loop is equivalent). In each of these separate approaches, the baseline wander compensation has been less than satisfactory. For example, detector delay plays an important role and thus has created significant problems in approaches (1) and (3).
With respect to method (1), in one particular exemplary application of an error feedback loop, the worst-case baseline wander signal has a frequency of 0.001 ftbg (sample frequency with a 1000 bit period), which requires at least 20 dB attenuation for good performance. Accordingly, the gain gblw has to be at least 0.01 ftbg and the loop is stable as long as the Viterbi delay is 25 T or less. Unfortunately as the system performance approaches the critical delay, the loop exhibits huge overshoot, and performance suffers.
In an example of method (2), a linear equalization device uses the inverse of the HPF transfer function H(z) that results in:
      1    -                  (                  1          -          Δ                )            ⁢              z                  -          1                          1    -          z              -        1            where
  Δ  =            2      ⁢      π      ⁢                          ⁢      α              1      +              π        ⁢                                  ⁢        α            and α is the fractional corner frequency. However, this system is not stable. The DC input gives unbounded output. Attempts to try moving a pole slightly inside the unit circle results in
            1      -                        (                      1            -            Δ                    )                ⁢                  z                      -            1                                      1      -                        (                      1            -            ɛ                    )                ⁢                  z                      -            1                                .This results in a huge boost at DC with Δ/ε which will have a negative performance impact. Accordingly, a linear equalization device with a pole close to the unit circle requires a large amount of precision. In this regard, the results obtained when using a linear equalization device have been unsatisfactory.
The use of method (3), a decision feedback equalization filter, to model the tail of the response after the Viterbi delay also does not have satisfactory results. In this regard, this filter does not work well with the high HPF cutoffs as the portion of the tail not cancelled is large. Accordingly, the decision feedback equalization filter has not been a viable solution.